In this talk I will present work on using coordinate-based neural networks for visualizing field-based data, e.g. data produced from large-scale numerical simulations or medical images. I will demonstrate that such a class of models are beneficial for visualization in three distinct ways: for the compression of fields, the reconstruction of fields from subsampled data, and for the efficient computation of derived visual quantities. A common theme in my work is the development of techniques that are informed by how fields are commonly visualized, namely, gradient-preserving compression, integration-based reconstruction of vector fields, and surrogate models of flow maps learned solely from vector fields.

Bio:Matthew Berger is an assistant professor in the Department of Computer Science, Vanderbilt University. He was a postdoctoral scholar at the University of Arizona from 2016 - 2018. He received his PhD in Computing from the University of Utah in 2013, as well as his MS and BS in Computer Science from Binghamton University in 2007 and 2005, respectively. His research interests are in data visualization and machine learning.